How to read a two-point lead in a poll

A poll puts one candidate at 48% and another at 46%. The difference is two percentage points in that sample. How much confidence should you put in the order? Start with the poll itself. Its field dates, question wording and sampling method can change what the result is useful for.

Who answered the question?

The American Association for Public Opinion Research asks pollsters to disclose the population they want to describe and how they selected or recruited the sample. A survey of all adults covers a wider group than a survey of registered voters. A likely-voter survey can narrow that group further using a model. If two polls use different groups, a different result could come from whom they include.

Check the field dates as well as the publication date. Interviews conducted over a week can include answers from before and after a major news event. A poll released on Friday may have finished interviewing on Tuesday.

Ask about the gap itself

Suppose the fictional survey lists a margin of sampling error of plus or minus three points for each candidate’s share in the full sample. The uncertainty around the difference between the two shares is a separate calculation. The answers come from the same sample and the two estimates are related. The published margin for one share alone cannot establish that a two-point lead holds in the population.

Some online panels recruit volunteers rather than select people with known probabilities. Their reports may use a model-based uncertainty measure. Check how the pollster describes that measure and the assumptions it uses. A margin for the entire sample also needs care when the headline concerns a much smaller subgroup.

Check the rest of the result

Ask for the exact question and response options. If the 48% and 46% came from a multiple-choice question, the remaining respondents might have selected another candidate, said they were undecided or declined to answer. Two polls can appear to disagree because one presents the full sample and the other recalculates shares among people who expressed a choice.

When comparing results over time, look first for releases with the same question, population and approach. Keep a dated note of each poll. Seeing the same movement in several comparable surveys is more informative than comparing a single new figure against the most dramatic earlier one.

• Record who was surveyed and when.
• Read the question and the treatment of undecided respondents.
• Look for uncertainty on the lead or describe it plainly as the difference in the published estimates.